A novel approach to action recognition in video based on the analysis of optical flow is presented. Properties useful for are captured using only empirical covariance matrix a bag features such as velocity, gradient, and divergence. The feature low-dimensional representation dynamics that belongs Riemannian manifold. manifold matrices transformed into vector space symmetric under logarithm mapping. log-covariance test segment approximated by sparse linear combination training segments...

We propose a general framework for fast and accurate recognition of actions in video using empirical covariance matrices features. A dense set spatio-temporal feature vectors are computed from to provide localized description the action, subsequently aggregated an matrix compactly represent action. Two supervised learning methods action developed matrices. Common both is transformation classification problem closed convex cone into equivalent vector space symmetric via logarithm. The first...

Action recognition is a challenging problem in video analytics due to event complexity, variations imaging conditions, and intra- inter-individual action-variability. Central these challenges the way one models actions video, i.e., action representation. In this paper, an viewed as temporal sequence of local shape-deformations centroid-centered object silhouettes, shape silhouette tunnel. Each represented by empirical covariance matrix set 13-dimensional normalized geometric feature vectors...

Aliasing is a common artifact in low-resolution (LR) images generated by downsampling process. Recovering the original high-resolution image from its LR counterpart while at same time removing aliasing artifacts challenging interpolation problem. Since natural normally contains redundant similar patches, values of missing pixels can be available texture-relevant pixels. Based on this, we propose an iterative multiscale semilocal method that effectively address The proposed estimates each...

Latest trend in image sensor technology allowing submicron pixel size for high-end mobile devices comes at very high resolutions and with irregularly sampled Quad Bayer color filter array (CFA). Sustaining quality becomes a challenge the signal processor (ISP), namely demosaicing. Inspired by success of deep learning approach to standard demosaicing, we aim investigate how artifacts-prone can benefit from it. We found that deeper networks are capable improve reduce artifacts; however, be...

In this paper, we investigate the features of spatiotemporal change in fractional vegetation cover (FVC) throughout Yellow River Basin between 2000 and 2022 identify driving factors behind using MODIS normalized difference index (NDVI) as a data source. On that basis, our research involves trend analysis center migration model to examine correlation changes various factors, such climates, topographies, soils, human activities. way, aim uncover how contributed reduction suspended sediment...

Learning-based super-resolution (SR) methods are popular in many applications recently. In these methods, the high-frequency details usually found or combined through patch matching from training database. However, representation ability of small is limited and it difficult to guarantee that super-resolved image best under global view. To this end, authors propose a statistical learning method for SR with both local constraints. More specifically, they introduce mixture model into maximum...

The prevention and control of soil erosion through water conservation measures is crucial. It imperative to accurately quickly extract information on these in order understand how their configuration affects the runoff sediment yield process. In this investigation, intelligent interpretation algorithms deep learning semantic segmentation models pertinent remote sensing imagery were examined scrutinized. Our objective was enhance accuracy automation by employing an advanced learning-based...

Rolling shutter effect commonly exists in a video camera or mobile phone equipped with complementary metal-oxide semiconductor sensor, caused by row-by-row exposure mechanism. As resolution both spatial and temporal domains increases dramatically, removing rolling fast effectively becomes challenging problem, especially for devices limited hardware resources. We propose method to compensate effect, which uses piecewise quadratic function approximate trajectory. The duration of each segment...

Action recognition is an important but challenging problem in video analytics with a number of solutions proposed to date. However, even if reliable model for action representation identified and accurate metric comparing actions developed, it still unclear how many frames should the comparison apply. In this paper, we develop method detect when change, i.e., temporal boundaries actions, without classifying actions. We use silhouette-based framework comparison, both centered around...

In language testing literature, the lexical profiles issue has been extensively discussed when examining quality of reading texts in high-stakes tests. The interpretation and use profiles, however, have lacking a point reference (i.e., benchmarks). Therefore, this study attempts to establish benchmarks for test China – National Matriculation English Test (NMET). To elicit sufficient samples, corpus 909 NMET was constructed. Based on corpus, two stages were employed. Firstly, screened through...

Motion deblurring techniques have played important roles in many image processing applications. In this paper, a new algorithm is proposed for motion from single image. The introduces an anisotropic Patiral Differential Equation (PDE) method latent prediction and employs adaptive optimization model the blind deconvolution iterative step. Furthermore, order to solve spatially-variant blur problems, we also propose Saliency-based Deblurring (SD) approach by introducing saliency extraction...

With large amount of industrial data available, data-driven anomaly detection methods have been widely used for ensuring safety and preventing economic losses. Among them, one-class support vector machine (OCSVM) is an effective unsupervised method detecting abnormal points by establishing a decision boundary in the kernel space has obtained much attention research field recent years. Through solving convex quadratic programming problem, OCSVM obtains classifier with maximum distance to...

In patch based face super-resolution method, the size is usually very small, and neighbor patchespsila relationship via overlapped regions only to keep smoothness of reconstructed high-resolution image, so prior not always strong enough regularize when observed low-resolution image lose facial structure information. We propose use Gaussian Mixture Model(GMM) learn embedded between un-overlapped patches. This approach, which has never been used before, works as a potential function in...

In learning based single image super-resolution (SR) approach, the super-resolved are usually found or combined from training database through patch matching. But because representation ability of small is limited, it difficult to guarantee that best under global view. To tackle this problem, we propose a statistical method for SR with both and local constraints. Firstly, use maximum posteriori (MAP) estimation learned priors by fields experts (FoE) model, regularize globally guided priors....

Image interpolation addresses the problem of obtaining high resolution (HR) images from its low (LR) counterparts. For observed LR with aliasing artifacts caused by undersampling, commonly used methods cannot recover HR well, and may often interpolate over-fitting artifacts. In this paper, based on observation that natural normally have redundant similar patches, a new patch-synthesis-based method is proposed for image interpolation. method, an inference Markov chain adopted to select best...